Hierarchy is Detrimental for Human Cooperation

Originally posted at Nature.com


Studies of animal behavior consistently demonstrate that the social environment impacts cooperation, yet the effect of social dynamics has been largely excluded from studies of human cooperation. Here, we introduce a novel approach inspired by nonhuman primate research to address how social hierarchies impact human cooperation. Participants competed to earn hierarchy positions and then could cooperate with another individual in the hierarchy by investing in a common effort. Cooperation was achieved if the combined investments exceeded a threshold, and the higher ranked individual distributed the spoils unless control was contested by the partner. Compared to a condition lacking hierarchy, cooperation declined in the presence of a hierarchy due to a decrease in investment by lower ranked individuals. Furthermore, hierarchy was detrimental to cooperation regardless of whether it was earned or arbitrary. These findings mirror results from nonhuman primates and demonstrate that hierarchies are detrimental to cooperation. However, these results deviate from nonhuman primate findings by demonstrating that human behavior is responsive to changing hierarchical structures and suggests partnership dynamics that may improve cooperation. This work introduces a controlled way to investigate the social influences on human behavior, and demonstrates the evolutionary continuity of human behavior with other primate species.


Determining the conditions that facilitate cooperation in humans has been a challenge embraced by many disciplines; evolutionary biologists, psychologists, and social scientists have been attempting to tackle this question for decades1,2,3. Understanding when cooperation flourishes is of both theoretical and practical interest as our species faces environmental and societal challenges that may only be solved by working together4,5. Cooperation has been defined in many ways6; here we are referring to cases in which two or more individuals work together to achieve a common goal7. While this form of cooperation is not altruistic (nobody necessarily incurs a cost for cooperating), choosing with whom to cooperate and under what conditions to invest limited resources into cooperation poses a significant challenge for our species and others8,9,10.

The effect of the social environment on cooperation has received attention in studies of nonhuman animal behavior but has been largely overlooked in human research. Research with animals in the wild and under controlled conditions in captivity has consistently shown that social dynamics, and specifically the nature of the dominance hierarchy, has a large impact on cooperative outcomes9,11,12,13,14,15,16,17,18,19. Although variable in form, every animal society has some form of dominance hierarchy20,21. Hierarchy is defined as priority of access to resources and probability of winning competitive encounters22, and reflects underlying assymetries in power. A hierarchy can be characterized in terms of linearity and steepness22, with the former providing information about the degree of transitivity between individuals and the latter indicating the extent to which individuals differ from each other in winning encounters or accessing resources. Among nonhuman primates, it has been demonsrated repeatedly that the characteristics of dominance hierarchies impact cooperative outcomes, with steep and linear hierarchies being associated with decreased cooperation. For example, experiments have shown that cooperation is impeded among chimpanzees living in steep and linear hierarchies16,23, whereas it emerges more easily among species with more relaxed hierarchies such as cottontop tamarins15,16,17.

A great deal of research has focused on human cooperative behavior24, but these experiments have primarily been conducted with anonymous participants25,26, leaving the influence of social relationships on cooperation largely overlooked27. Although the influence of hierarchy on cooperation has rarely been examined, researchers have considered hierarchy’s influence on economic issues such as market entry28, bargaining29,30, and learning31. Other work has investigated how disproportionate power in sanctioning influences cooperation32, and both empirical33,34 and modeling35,36 investigations have been directed at how group status impacts cooperation and competition with other groups. In the current study, we hypothesize that social relationships, and specifically hierarchical relationships reflecting power assymetries between individuals, will have a negative impact on human cooperation as it does in our nonhuman primate relatives. In order to test the effect of social hierarchy on cooperation, we present participants with a task inspired by nonhuman primate research in which two individuals of known social rank are presented with the opportunity to invest in a cooperative task, and, if a threshold of investment is met and cooperation is achieved, the higher ranking of the two investors controls the distribution of the resource16,37,38. To investigate how human cooperation is impacted by the presence of a social hierarchy, we compare cooperative success in the presence of a hierarchy (with both earned and arbitrarily assigned ranks) to success in a condition when hierarchy is absent.

Experimental design

In order to probe the effects of hierarchy on human cooperation, we adapted a classic mutualistic cooperation task employed in nonhuman primate studies15,16,39,40 for use with human participants. In this task, two individuals have the opportunity to work together to obtain some benefit that is not pre-divided for the players. In the present study, the only social feature we manipulated was hierarchy, which allowed us to isolate this effect while preventing confounding effects arising from other social relations. To generate the hierarchies, we asked groups of ten anonymous participants to carry out multiple unrelated tasks on a computer (see Methods Summary below and Supplemental Information for a full description), and ranked them according to their overall performance. Subsequently, participants were arranged in pairs and participants were informed of their own rank and the rank of the person with whom they were paired. In the cooperation phase, both players in the pair were assigned 20 units and given the option to contribute simultaneously any number of these units to a common pot, unaware of the partner’s contribution. ECUs not contributed to the common pot were not lost to the player but could be obtained in the final payout. If this pot reached or exceeded 20 units, then the pot doubled to 40 units for the splitting phase, otherwise they obtained nothing (i.e., their contributions to the pot were lost). Players were informed only whether the threshold was reached or not, and did not know a posteriori the amount contributed by their partner.

The splitting phase was implemented with an ultimatum game. The higher-ranked individual proposed how to split the 40-unit pot, and if the proposal was accepted by the lower-ranked person, each would receive the stipulated amount. If the lower-ranked person rejected the higher-ranked person’s proposal, s/he could attempt to compete and a lottery would assign the 40 units to one of the two members of the pair with probability proportional to their rank (which was known to the participants). Given that hierarchy is defined as priority of access to resources and the probability of winning competitive encounters22, these methods were designed to maintain the impact of hierarchy throughout the experiment.

Players played nine rounds, interacting with every other player in the group so all possible rank differences could be explored. Each player experienced one of three different treatments: the “earned hierarchy condition” (described above), a “random hierarchy condition” in which the ranking was assigned randomly without performing any task, or a control condition with no hierarchy. In the control condition, when the proponent’s offer was rejected, the whole pot was assigned to one of the players with equal probability.


In what follows, we analyze the experimental results using mixed-effects models, the appropriate technique when individuals are embedded within a group and are paired with each other across several interactions (and hence are not independent). This approach, combined with random selection of the individuals representing each dyad (when dyads were the proper level of analysis) and a bootstrap resampling technique, allows us to establish significance with a large degree of accuracy (see Supplemental Information).

First, we look at if (and how) the existence of a hierarchy affected success in the cooperation phase. Figure 1A shows that success (reaching or exceeding 20 units collectively) in the control condition was significantly more frequent than that of the hierarchically organized groups (95% CI for the coefficient, [0.14, 0.45]), indicating that participants were more prone to cooperation in the treatment lacking hierarchy. Interestingly, there was no difference between the random and earned hierarchy conditions, indicating that whether rank arose from personal performance or was randomly assigned did not affect cooperation success. Therefore, we pooled the two hierarchical treatments in subsequent analyses. In the control condition the average contribution to the pot was significantly greater than in the hierarchical treatments (95% CI for the coefficient, [0.23, 0.58]), cf. Fig. 1B) leading to more successful cooperation events in the absence of a hierarchy. We also found an interaction between the presence or absence of hierarchy and round of play, which emerged from the fact that the difference in cooperative success between the hierarchy and no hierarchy conditions increased as the experiment proceeded (95% CI for the coefficient, [0.007, 0.116], cf. Fig. 1C). Finally, we observed that there was a positive correlation between rank and total earnings in the experiment (see Supplemental Information), i.e., higher ranked participants received larger payments than lower ranked ones. Therefore, in this experiment we did in fact see that higher ranked individuals obtained more resources.

Figure 1: Success and contributions decrease in hierarchically organized groups, irrespective of the origin of the ranking.

Figure 1

(A) The mean number of succesful instances of cooperation in the control condition in which there was no hierarchy was significantly higher than both hierarchical conditions. The maximum number of cooperative successes was 9. (B) The mean player contribution in the cooperative task with and without hierarchy. (C) Mean cooperation success as a function of round and hierarchy condition shows that there is a small but significant interaction among these variables, which might arise from the fact that the differences between the two treatments appear to increase as the experiment progresses. Colors correspond to the three types of hierarchical treatment as indicated in the plot.

Why does cooperation suffer in the presence of a hierarchy? While the contributions from both partners are similar in the condition with no hierarchy, they differ clearly in the two hierarchical treatments (Fig. 2A). The main reason for the decrease in successful cooperation events can be traced to diminishing contributions by lower ranked individuals (Fig. 2B). Indeed, in this respect, our analysis shows that there is a significant difference in contributions between higher and lower ranked individuals (95% CI for the coefficient, [−1.65,−0.79]) and a significant interaction of rank and round (95% CI for the coefficient, [−0.30, −0.10]). Figure 2B shows that higher ranked participants increase their contributions as the experiment procedes while lower ranked participants decrease their contributions. Thus, it is the lack of contribution from the lower ranked players that leads to more cooperation failures. Figure 3 shows that, in the hierarchy conditions, lower ranking individuals contribute little in unsuccessful attempts (when the 20 unit minimum is not reached) compared to higher ranking individuals, and when cooperation is achieved, lower ranking individuals’ contributions barely surpass ten units. In view of this evidence, it appears that both individuals realized the consequences of their rankings on their potential earnings. Accordingly, lower ranked participants responded by reducing their investment, whereas higher ranked participants anticipated the reluctance of their partner and invested more in an attempt to rescue cooperation, but often not enough to be successful. In fact, when we examined whether the magnitude of the rank difference predicted cooperation investments, we found that the amount contributed by lower ranked participants increases (and the amount contributed by higher ranked participants decreases) as the rank difference became smaller, i.e., as the chances of receiving the whole pot by chance increased (95% CI for the coefficient, [−0.47, −0.25]) (cf. Fig. 2C).

Figure 2: Contributions decrease in the hierarchy condition for lower ranking partners, and are predicted by the rank difference.

Figure 2

(A) The contributions by the higher and lower ranked partners in the three experimental conditions, (B) the contributions by the higher and lower ranked partners in the hierarchy and non-hierarchy treatments across rounds, (C) the contribution as a function of the rank difference between the two partners of the dyad. Negative numbers correspond to higher ranking positions, e.g., −8 indicates the focal player (whose contribution we are evaluating), was ranked 1 and her partner was ranked 9).

Figure 3: Contributions differ markedly in the cases when cooperation is or is not achieved.

Figure 3

(A) Mean player contribution in the cooperative task when cooperation fails (red) and when it succeeds (blue) for the higher ranked player in the three experimental conditions. (B) The same for the lower ranked player. Note that when there is no hierarchy the behavior of both types of player is the same.

Let us now move on to the behavior of players in the ultimatum game with an (hierarchy-based) outside option41, or the “splitting phase”. In the absence of hierarchy, we observed that proponents offered on average 25% less than what receivers were willing to accept. When hierarchy was introduced, this difference was again observed, but proponents offered lower amounts for higher rank differences and receivers stated a lower minimum amount they would accept (Fig. 4). On the other hand, our analysis shows that both offers and expectations are independent of the investments made in the cooperation phase. This finding is further supported by the results of two additional treatments in which the cooperation phase was omitted and players proceeded directly from the hierarchy formation stage, be it earned or random, to the splitting phase. We did not observe any significant differences in the amounts offered and expected between these treatments and those in which there was a cooperation stage (see Supplementary Information). Such a result may be explained by a similar feeling of entitlement in the splitting phase regardless of whether this phase followed successful cooperation.

via Nature Creative Commons License 4.0

Read More: http://www.nature.com/articles/srep18634